Nianyi Li
Clemson University, SC·Computer Science·Early-career
Computer vision lets machines turn camera images into an understanding of the 3D world. Most methods, however, were built for orderly places such as indoor rooms and city streets.
- Fields
Computer Science
Engineering
- Topics
computer vision
3D vision
3D reconstruction
- Grants
- 2 known grants, through May 2030
What they work on
- Build a multi-sensor imaging system that combines RGB cameras, depth sensors, LiDAR and event cameras to capture diverse data from unstructured outdoor environments.
- Develop transformer-based neural networks that merge short- and long-range sensor data to extract and match visual features that stay consistent across space and time in off-road scenes.
- Create adaptive 3D reconstruction methods based on hierarchical Gaussian Splatting, which represents a scene as many small 3D blobs, to model large terrain under varying lighting and weather.
- Design real-time tracking algorithms that handle blocked views, clutter and moving objects, and benchmark them against existing methods in real-world environments.
- Separate object motion from turbulence-induced motion in long-range images using neural fields and physics-based velocity fields, so that depth and fine color detail can be recovered.
- Estimate turbulence strength and flow in air and water from ordinary images with deep neural networks, and use these motion fields to guide intelligent foveation for long-range vision tasks.
More about the research
They often fail on rough off-road terrain, in disaster zones, or when images are taken from far away through turbulent air or water. This research asks how to make machine vision reliable under these hard conditions. One project combines several kinds of sensors to understand messy outdoor scenes. The other turns the distortions caused by turbulence into useful information for long-range imaging.
Why it matters
More dependable 3D perception could help self-driving vehicles cross rough terrain, help search-and-rescue drones spot obstacles in disaster areas, and improve environmental monitoring of large outdoor scenes. Methods that correct for, and even exploit, turbulence could improve drone navigation and long-range object tracking and detection. Public datasets from both projects could support further machine-learning research.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Jinwei Ye
George Mason University, VA·Computer Science·Early-career
Model how the polarization of light changes when it reflects off complex surfaces, using physics-based layered material models and a shared high-precision polarimetric reflectance dataset.
Noah Snavely
Cornell University, NY·Computer Science
Develop compositional implicit neural scene representations that pair a network's ability to capture fine detail with physical models of scene structure, such as light fields.
Rui Yu
University of Louisville, KY·Computer Science
Extract a labeled structural skeleton from single noisy images by fitting geometric primitives such as planes and cylinders to monocular depth estimates, using large vision-language models to assign meaning.
The school
Clemson University: 19 labs on PI Pages, $339M a year on research, 297 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.